[diffusion] refactor: refactor fuse qkv with QKVParallelLinear linear (#15090)

Co-authored-by: Mick <mickjagger19@icloud.com>
This commit is contained in:
Xiaoyu Zhang
2025-12-15 00:33:29 +08:00
committed by GitHub
co-authored by Mick
parent d277a86dea
commit 64b5c3ab90
9 changed files with 118 additions and 285 deletions
@@ -584,7 +584,7 @@
"48": 247.48,
"49": 247.54
},
"expected_e2e_ms": 16563.83,
"expected_e2e_ms": 18382.19,
"expected_avg_denoise_ms": 260.76,
"expected_median_denoise_ms": 247.84
},
@@ -706,7 +706,7 @@
"TimestepPreparationStage": 2.9,
"LatentPreparationStage": 1.25,
"ImageVAEEncodingStage": 1655.89,
"DenoisingStage": 100544.98,
"DenoisingStage": 106972.82,
"DecodingStage": 1355.52,
"per_frame_generation": null
},
@@ -753,7 +753,7 @@
"39": 1599.78
},
"expected_e2e_ms": 123182.9887,
"expected_avg_denoise_ms": 2513.52,
"expected_avg_denoise_ms": 2831.00,
"expected_median_denoise_ms": 1600.09
},
"wan2_1_i2v_14b_480P_2gpu": {